Muscle memory is a big factor in fencing, but I doubt you are interested in simulating that. What I advise is making certain moves follow naturally from other moves. I was always trained to riposte after every parry. If you just stop his attack and don''t come back at him, you''re giving the bout away. SO, when your little guy parries a cut, he should be able to respond immediately. I fenced saber, and when I parried in line 4, I''d nearly always riposte in line 5, because unless the guy''s attack was half-hearted, or a feint, he''d never recover from his lunge in time to parry 5. If he was baiting me, and I suspected it, I''d go 3, and get him under his sword. There''s a huge psychological element in fencing, a sort of paper-rock-scissors to the tenth power, but it shouldn''t be impossible to at least make something that looks like it. Although it does deprive the sport of some of it''s *sniff* nobility.

I don''t know if you fence or not, but your time model is probably inadequate. More important than the speed of the move is the reaction time of the fencer. If I see that attack coming, I can stop it no problem, and likewise for the other guy. Any defensive move is faster to execute than any offensive move, and so it those are your two variables then you''ll have a very rapid clang-clang sort of phrase. But speed is still important. I can attack 3 and then parry his riposte in five without much difficulty, even if I''m a little slow on the uptake, because that''s a very quick move. But going from a cut in 4 to a parry in five requires precognition. You have to know you''re going to do it even before you finish that cut, or else your blade will never make it. That''s how you string them together.

A deep lunge with an attack in 4 prety much dooms you unless you hit the guy. So you never start out with one of those. Assign a percentage to it. A low one. A straight attack in 5 is quick and recovery is equally swift, so it''s a good opener. 3 is good for a change-up, but a little slower than 5. If your attack in 3 is parried, you can bet your ass that the guy is going to riposte, and most likely in 5. So you can actually do that one with muscle memory before his attack begins. If he sees it, though, he''ll go 3 to avoid your parry. Tricky. Action and reaction blend together in an exchange, but usually either a point is scored or the fencers disengage after three or so swings. Only once did I actually go the length of the strip with parry-riposte back and forth, and the other guy got the touch anyway. Bastard. usually it''s clang-clang-clang-"I''m out of here" and then you re-engage and do it again.

Things like distance and footwork are instrumental as well. But if you were going balls-out as a simulator, you''d never get your assignment done. You may or may not want to throw in beat-attacks, feints, remizes (sp?) point-in-line, vertushkas (sp?) and all the other fancineses of the sport. For your purposes, elementary moves and strategies would be most practical. Attack, parry, riposte, disengage and repeat.

Also, use epee rules. Foil and saber both feature right-of-way, which is a very complex and subjective scoring element, basically so that you can''t both kill each other at the same time. Epee allows simultaneous hits, and so you don''t have to worry about whose arm extended first or anything like that. However, epee also allows touches all over the body, which means more lines in which to attack, defend, etc.

Good luck with your project.

quote:Original post by TechnoGoth
Complexity – the complexity of the move, the more complex the move the more likely it will succeed.

it doesn´t work that way - a move only succeeds if it has the right degree of complexity.
That means that when the attackers intended action consists of 3 elements (feint - feint - hit), but the defenders of only two (parade - riposte) then the defender will score the hit.

also, I agree about the fatigue value. while fencing is physically very demanding you won´t run out of steam during a bout, the only time when that kind of thing factors is when you have to go up against multiple opponents in quick succession, or do a day-long tournament.

as for making combat more sequential - add the possiblity of one or both of the fighters moving out of range again.

if you want a gradual buildup until a hit occurs I´d play around with these variables:

pressure: don´t just calculate success/failure for an action, but by which degree it does so. if for instance a defensive action succeeds by a lot, the defender may regain the initiative and carry some bonus from the previous action into his counterattack.
increasing the distance to the opponent decreases pressure for both.
a fencer moving into the warning zone (the respective end areas of the strip) pressure would also increase, since he cannot move back indefinitely.

reaction to moves in progress: let the agents react to moves that are just beginning, especially for Epee this is very important as there´s a lot of action without the intention to hit there (testing the opponents reactions and reflexes, building a rhythm etc.)

While it is *extremely* simplied, I would highly suggest that you try "researching" (playing) the original Prince of Persia (the Apple II version would obviously be hard to get, but the old PC version can actually run on Windows XP). It successfully merged easy to learn, skill based (not button mashing) and elegant sword fighting.

Well, with 4 attacks and four (presumably corresponding) defenses, I think you''d have little difficulty making a machine learning algorithm. Since fancing is little more than a graphical shell for this algorithm, an actual understanding of the sport is superfluous. I think that the key ingredient you need is the ability to give the impression that attack #1 is going to take place, but actually perform attack #2. That''s a feint, after all. Letting the defender record and analyze data regarding frequency of certain events, success rates of its own actions, and patterns in its opponent''s behavior would be a simple mathematical excercise, and could be performed admirably by any respectable machine. You should have little difficulty with your assignment.